Fixed calculation of norm for per-atom rank-2 tensors in get_parity_data
This MR fixes the norm selection for atomic_v of the get_parity_data function when it represents a
per-atom polarizability (6-component reduced Voigt tensor) rather than a
per-atom dipole (3-component vector). The norm was previously computed as a
naive Euclidean norm over the raw 6 Voigt components, instead of expanding to
the full symmetric 3x3 tensor and taking the Frobenius norm (as already done
for virial/stress). This under-weighted the off-diagonal (xy, xz,
yz) contributions, since they each appear twice in the full tensor.
The 3-component dipole-like case is unaffected.
Testing
Fixed test_get_parity_data_norm, which previously mirrored the same bug in
its expected-value calculation and so could not catch it. Added two new
tests covering the 3-component and 6-component atomic_v norm cases
separately, the latter asserting the corrected result differs from the old,
incorrect one.